Why Did the Ant Ling Team Redesign SwiGLU? The Birth of the PowLU Activation Function
Original title:我们为什么要重新设计 SwiGLU?聊一聊 PowLU 的诞生始末
AI summary
The Ant Ling team proposed a new activation function, PowLU, to solve the Loss Spike and training collapse problems caused by SwiGLU's quadratic growth amplifying outliers in low-precision training such as FP8.
Quick Facts
- Published
- Updated
- Source
- 公众号:蚂蚁百灵(Ling)
- Category
- Papers & Research
- AI score
- 81
用幂函数替代指数项以压制输出范围,避免了 SwiGLU 截断带来的信息丢失,为低精度训练提供了一条已通过千亿模型验证的稳定激活函数路径。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Why Did the Ant Ling Team Redesign SwiGLU? The Birth of the PowLU Activation Function. May 29, 2026. https://new.nat6.net/en/news/cmsdu2qga010jrofz571wp958/